Bloomberg chart showing just how much RAM datacenter GPUs take.
Nvidia’s H100 carried 80GB of GPU memory, H200 moved to 141GB, Blackwell moved to 192GB class parts, and GB300 Blackwell Ultra reaches 288GB of HBM3e per GPU.
A 72-GPU rack multiplies that into a memory wall big enough to change supplier behavior, because AI servers pay premium prices for capacity and bandwidth that make everyday device memory less attractive
FT: Apple is asking Washington for permission to buy DRAM from CXMT, a blacklisted Chinese supplier, because AI server demand has made ordinary device memory pa...